Recently, an image monitoring system or an image black box for automobiles, or the like is used for detection or prevention of accidents. Further, study is under progress for providing lane deviation or car collision warning by extracting lanes and a vehicle ahead from the image acquired by a video camera regarding a safety automobile of high technology.
In order to improve the performance of a computer vision application system or image process, a clean input image is needed. Particularly, when detecting or estimating an object or using the edge information of the image, the cleaner image will bring a better result.
However, in case of image taken outdoors, the brightness and the color reflected from an object are mixed with particles in the air. Thus, its original color and brightness contrast are not provided, especially, in case that there are big particles such as a fog and smoke in the air,
As for the conventional method for improving an image which includes fog and etc., there is Korean Patent Publication No.10-2010-0021952 (hereinafter, ‘prior art’) in addition to a plurality of applications published and disclosed.
The method according to the prior art comprises a step of receiving the first luminance image of an image including Airlight, and producing a map of atmosphere Airlight based on the ratio of the standard deviation and the average of the first luminance image; and a step of outputting the second luminance image from which the atmosphere Airlight is removed by subtracting a map of atmosphere Airlight produced from the first luminance image. However, in the prior art a fogged image process is processed by a block unit, but not by a pixel unit.
On the other hand, a variety of methods for estimating a clean image by removing fog from a fogged image have been proposed. In the early stage, a method using a number of images or additional information besides to an image was proposed to restore a fogged image to a fog-eliminated image. As the method using a number of images, the method using polarization [1, 2] was proposed. This method acquires two or more images taken with respectively different polarization filter installed in precisely same location, and the depth value is calculated by using a method for measuring the polarized amount and the fog is removed by using it. However these methods provide a very good image in result, there is a strong limitation that respectively different polarization filters have to be used in the same location. In a method [3, 4, 5] using simply a number of images without using polarization filter, fog value and depth information are obtained from two images taken under different weather environment to remove the fog. Kopf, et al. [6] proposed a method for removing fog by using depth information of an image instead of using a number of images, and the fog was removed by obtaining depth or texture information by using GPS information embedded in a camera and assuming concentration (density) of fog value is depth information.
The methods for removing fog by using a number of images or using a single image with additional information have a defect that they cannot be adapted to an image taken by a dynamically-moving camera, because they need to secure image data in various conditions. Thus recently a method for removing fog by using a single image is being studied.
Tan [8] proposed a method for removing fog by increasing brightness contrast. That is, fog was removed by using the characteristic that a clean image without fog had higher edge strength than a fogged image and fog value didn't change rapidly. In this method, the brightness contrast is highly improved, thus the shape and structure of an image are advantageously revealed. But excessive increase of contrast may cause saturation. And halo effect may occur in the section where the depth information is largely different.
Fattal [9] proposed a method for restoring a fog-eliminated image by measuring the image reflection ratio through assumption that reflection ratio measured within a constant image area has always the same vector direction.
He et al. [10] used the characteristic that a clean image has a higher chroma of color than a fogged image, and proposed a method which removes fog by the observation result that a pixel with high color sharpness of a clean image without fog has a very low channel value of one of R, G, B values, thus a color image without fog has a pixel with very low channel value in a certain area. However, in case only a luminance image is used because RGB color is used in the conventional method using only a single image, fog-elimination performance is degraded. And as a large size of filter is used, halo effect occurs and a large calculation amount is required to refine transmission rate. Thus there is difficulty in real time processing.
Tarel et al. [11] proposed a fog elimination method using a median filter in order to improve the calculation speed, however, if using a large size of a median filter, there is a disadvantage that the calculation speed is decelerated, and a back light effect can occur.
[1] Y. Y. Schechner, S. G. Narasimhan, and S. K. Nayar, “Instant dehazing of images using polarization,” in Proc. CVPR, pp. 1984-1991, Hawaii, USA, Dec. 2001.
[2] S. Shwartz, E. Namer, and Y. Y. Schechner, “Blind haze separation,” in Proc. CVPR, pp. 1984-1991, New York, USA, Oct. 2006.
[3] S. G. Narasimhan and S. K. Nayar, “Chromatic framework for vision in bad weather,” in Proc. CVPR, pp. 598-605, SC, USA, June 2000.
[4] S. G. Narasimhan and S. K. Nayar, “Contrast restoration of weather degraded images,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 25, no. 6, pp. 713-724, June 2003.
[5] S. K. Nayar and S. G. Narasimhan, “Vision in bad weather,” in Proc. ICCV, pp. 820-827, Corfu, Greece, Sep. 1999.
[6] J. Kopf, B. Neubert, B. Chen, M. Cohen, D. Cohen-Or, O. Deussen, M. Uyttendaele, and D. Lischinski, “Deep photo: Model-based photograph enhancement and viewing,” ACM Trans.
Graphics, vol. 27, no. 5, pp. 1-10, Dec. 2008.
[7] S. G. Narasimhan and S. K. Nayar, “Interactive deweathering of an image using physical models,” In Workshop on Color and Photometric Methods in Computer Vision, OCT. 2003
[8] R. Tan, “Visibility in bad weather from a single image,” in Proc CVPR, pp. 1-8, Alaska, USA, June 2008.
[9] R. Fattal, “Single image dehazing,” ACM Trans.
Graphics, vol. 27, no. 3, pp. 1-9, Aug. 2008.
[10] K. He, J. Sun, and X. Tang, “Single image haze removal using dark channel prior,” in Proc. CVPR, pp. 1956-1963, Miami, USA, June 2009.
[11] Tarel, Jean-Philippe; Hautiere, Nicolas; “Fast visibility restoration from a single color or gray level image,” Computer Vision, 2009 IEEE 12th International Conference on, pp.2201-2208, Sept. 2009.